Bernstein-type exponential inequalities in survey sampling: conditional Poisson sampling schemes (Q2325384)

Results concerning upper bounds for the probability that the sum of independent random variables is larger than its expectation by a given value \(t\), are well known in literature. However, for complex survey designs, when there is a dependence structure between the sample values, similar bounds are obtained only for certain cases such as SRSWOR, Poisson sampling etc. The main purpose of this paper is to extend tail bounds for rejective sampling viewed as conditional Poisson given the sample size. The probability of deviation is expressed as the ratio of two quantities \(N\) and \(D\). Here, \(N\) is the joint probability that a Poisson sampling based estimator of the population total is larger than a given value \(t\) and the size of the cardinality of the Poisson sample is the sample size n of the rejective plan, while \(D\) is the probability that the Poisson sample size is \(n\). The preliminary results concern tail bounds for Poisson sampling, followed by inequalities for negatively associated random variables. The main theorems provide exponential inequalities for rejective sampling. A numerical illustration related to the bounds established in the theorems is also given towards the end. Finally, the authors explain how extensions to control tail probabilities for more general sampling designs can be obtained. An appendix of nine pages gives technical proofs of the main theorems.